Distortion-Invariant Target Recognition Based on Multi channel Joint Transform Correlator

被引：0

作者：

Lin Chao ^{[1
]}

Han Yanli ^{[1
]}

Lou Shuli ^{[2
]}

Liu Pei ^{[1
]}

Zhang Wenlong ^{[3
]}

Yang Zikang ^{[3
]}

机构：

[1] Naval Aviat Univ, Sch Aviat Operat & Support, Yantai 264000, Shandong, Peoples R China

[2] Yantai Univ, Sch Optoelect Informat Sci & Technol, Yantai 264000, Shandong, Peoples R China

[3] Unit 92485 PLA, Dalian 116041, Liaoning, Peoples R China

来源：

CHINESE JOURNAL OF LASERS-ZHONGGUO JIGUANG | 2022年 / 49卷 / 13期

关键词：

information processing; optical pattern recognition; multiple channeled joint transform correlator; synthetic discriminant function; distortion-invariant pattern recognition; IMPLEMENTATION; SCALE; IMAGE; PHASE;

D O I：

10.3788/CJL202249.1309001

中图分类号：

O43 [光学];

学科分类号：

070207 ; 0803 ;

摘要：

ObjectiveWiththeadventofthebigdataandintelligenceerasinformationsystemsrequireconsiderablyenhancedperformanceandlowenergycostsOpticalcomputingmaybecomethenext-generationcomputingplatformowingtoitsparallelprocessingcapabilityandhighbandwidthwithlowenergyconsumptionInpatternrecognitionapplicationslargeamountsofimagedatamustberapidlyprocessedTwotypesofopticalapproacheshavebeeninvestigatedforpatternrecognitionopticalneuralnetworkwhichcomprisestwosubclassesincludingsiliconphotonic-basedneuralnetworksandfree-space-basedopticalnetworkTheformerhasundergoneconsiderableadvancementsrecentlyowingtoimprovedfabricationcapabilityandnovelnetworkcomponentsbasedonopticssuchasmicroringresonatorsandMach-ZehnderinterferometersThelatteregdiffractiveneuralnetworksisalsoimportantparticularlyforcomputationalimaging-basedapplicationsHoweveropticalneuralnetwork-basedpatternrecognitionapproachesareimmatureowingtotheimplementationofnonlinearfunctionsPatternrecognitionapproachesfoundedonfree-space-basedopticalnetworksarehybridoptoelectroniccorrelatorsfarmorematurethanopticalneuralnetwork-basedonesThecorrelatorcanbecodesignedwithaneuralnetworktoserveasacoprocessertoprefiltersomeimagefeaturesforultrafastprocessingHoweverinconventionalopticalcorrelatorsboththespatialandspectralbandwidthsofsystemshavenotbeenefficientlyusedwhenperformingthecorrelationoperationHencetheinherentparallelprocessingcapabilityofopticscannotbefullyexploited. MethodsInourpreviousworkamultichanneljointtransformcorrelationmethodisproposedbasedonthecompressionandtranslationofjointtransformpowerspectrumtofullyutilizespatialandspectralbandwidthsandenhancetheparallelprocessingefficiencyandrecognitionaccuracyofopticalcorrelationsystemsIntheinputplaneofthisschemethescene imageandNnumbersofreferenceimagesareuploadedondifferentzonesoftheinputspatiallightmodulatorthenthephasemapsoptimizedusingtheiterativealgorithmaresuperimposedontotheimagesIntheFourierplaneinterferencebetweentheFourierspectraofsceneimagesandthoseofeverysinglereferenceimageoccursindifferentzonesoftheFourierplaneWhentherestrictionparameterinthephaseoptimizationalgorithmisappropriatedadjustednointerferenceoftheFourierspectraofthereferenceimagesisobservedConsequentlytheparallelprocessingofNchannelsisachievedwithoutcrosstalkTherelationbetweenthelocalizedpeakcluttermeanoftheFourierspectraofthepreferredphaseandthestandarddeviationofthecorrelationpeakpositionisanalyzedandusedasacriterionforpreferentialpreferredphasemaskselectionFurthermorethestandarddeviationofthecorrelationpeakpositionisobtainedforrecognitiontasksInthisstudywefocusondistortion-invariantpatternrecognitionbyintegratingthemultichanneljointtransformcorrelatorandthesyntheticdiscriminantfunctionFirstthefeasibilityofthelocalpeaktocluttermeanasaconstraintforpreferentialphaseselectionisanalyzedresultsindicatedthatthisfactorisnotappropriatewhenthesyntheticdiscriminantfunctionisusedHenceanewphaseselectioncriterion-knownasthevariationinthecorrelationpeakposition-isproposedtoobtainthepublicpreferredphasefortargetswithaspecificdistortionrangeFurthermoretheselectedphaseisusedinthemultichanneljointtransformcorrelatorwiththesyntheticdiscriminantfunctiontoachievedistortion-invariantpatternrecognitionThentodeterminethesystemperformanceintermsofthedistortionlevelthetoleranceofoursystemonthescaling-downofthesizeoftargetandtheincreaseinthenumberoftrainingimagesforthesyntheticdiscriminantfunctionareanalyzedFinallyconsideringthatthebackgroundmayvaryinrealapplicationswetakesuccessivevideoframesasvariedinputbackgroundsandanalyzethefeasibilityofourproposal. ResultsandDiscussionsResultsindicatethatundertheconsideredimagefilesizeandbackgroundcomplexitytheproposedmethodcanachievenine-channelparallelrecognitionFig8Forcorrectrecognitiontheminimumscalingdownfactoris06Fig11Whenthenumberofrotatedtrainingimagesisincreasedto9inthesyntheticdiscriminantfunctionacorrectrecognitioncanbeguaranteedFig13TherelationbetweentheminimumthresholdofphasetheoptimizationconstraintandthesynthesizedimagenumbersofSDFisobtainedforcalculatingthepreferredphasesTable1FurthermoreacorrectrecognitioncanbeguaranteedwhenthevaluesofhalfthepixelsinthebackgroundhavechangedFig16 ConclusionsHereinanoveldistortion-invariantpatternrecognitionmethodbasedonmultichanneljointtransformcorrelatorisproposedThelocalpeaktocluttermeanisshowntobeunsuitableandweproposeanewoptimizationcriterionknownasthevariationinthecorrelationpeakpositionwhichisfeasibleinthisproposalWeachievenine-channelpatternrecognitionwithin06--10timesofthescalingoftheimagesizeandrotationrangesof0 degrees--30 degrees 70 degrees--100 degrees 140 degrees--170 degrees 210 degrees--240 degrees and280 degrees--310 degrees TheupperlimitofthenumberofsynthesizedtrainingimagesisanalyzedwhichisnineinthisstudyMoreovertheproposedmethodcanmaintainitsperformancewhenthebackgroundisvariedwithinthevaluesofhalfitspixelindicatingrobustnesstobackgroundchangesTherecognitionspeedandaccuracyondistortionofthesystemareconsiderablyimprovedwithourproposalwhichwillbenefitthedevelopmentofpracticalmultichannelopticalcorrelators

引用

页数：16

共 29 条

[1] Distortion-invariant fringe-adjusted joint transform correlation [J].

Alam, MS ;

Chen, XW ;

Karim, MA .

APPLIED OPTICS, 1997, 36 (29) :7422-7427

[2] FRINGE-ADJUSTED JOINT TRANSFORM CORRELATION [J].

ALAM, MS ;

KARIM, MA .

APPLIED OPTICS, 1993, 32 (23) :4344-4350

[3] UNIFIED SYNTHETIC DISCRIMINANT FUNCTION COMPUTATIONAL FORMULATION [J].

CASASENT, D .

APPLIED OPTICS, 1984, 23 (10) :1620-1627

[4] Advances and Challenges of Optical Neural Networks [J].

Chen Hongwei ;

Yu Zhenming ;

Zhang Tian ;

Zang Yubin ;

Dan Yihang ;

Xu Kun .

CHINESE JOURNAL OF LASERS-ZHONGGUO JIGUANG, 2020, 47 (05)

[5] Implementation of a nonzero-order joint transform correlator using interferometric technique [J].

Cheng, CJ ;

Tu, HY .

OPTICAL REVIEW, 2002, 9 (05) :193-196

[6] MULTIOBJECT RECOGNITION IN A MULTICHANNEL JOINT-TRANSFORM CORRELATOR [J].

FENG, JH ;

CHIN, GF ;

WU, MX ;

YAN, SH ;

YAN, YB .

OPTICS LETTERS, 1995, 20 (01) :82-84

[7] Deep learning in photonics: introduction [J].

Gao, Li ;

Chai, Yang ;

Zibar, Darko ;

Yu, Zongfu .

PHOTONICS RESEARCH, 2021, 9 (08) :DLP1-DLP3

[8]

GERCHBERG RW, 1972, OPTIK, V35, P237

[9] OPTICAL-PATTERN RECOGNITION USING CIRCULAR HARMONIC EXPANSION [J].

HSU, YN ;

ARSENAULT, HH .

APPLIED OPTICS, 1982, 21 (22) :4016-4019

[10] JOINT TRANSFORM IMAGE CORRELATION USING A BINARY SPATIAL LIGHT-MODULATOR AT THE FOURIER PLANE [J].

JAVIDI, B ;

KUO, CJ .

APPLIED OPTICS, 1988, 27 (04) :663-665

← 1 2 3 →